For years, security teams have repeated the same mantra: “Turn on MFA, train your staff on phishing, and you’ll filter out most of the commodity noise.”
It’s good advice. But what happens when the attacker skips your users entirely and walks in through an enterprise integration you authorized three years ago?
That’s what Microsoft laid out yesterday in its report on Storm-0940, a group it describes as China-state aligned and active since 2021. The target wasn’t an unpatched firewall or an employee falling for a gift-card scam. The way in was ConnectWise’s Cloud Backup service, a tool countless managed service providers use to back up their customers’ Microsoft 365 environments.
Microsoft assesses that Storm-0940 took control of the identities of several ConnectWise cloud applications. Those applications already had trusted, high-privilege access to customer tenants through Entra ID. Once the attackers held them, the rules of the game changed completely.
The rise of agentic offensive tools
What made this campaign stand out wasn’t just the initial access. It was the speed. Storm-0940 ran its operation through an agentic AI framework Microsoft calls ARTEX. Instead of a human operator methodically typing commands into a terminal, coordinated AI agents handled targeting, authentication and reconnaissance.
Using the stolen app credentials, ARTEX attempted sign-ins to more than 2,250 Microsoft 365 tenants. In one slice of the activity, it logged more than 200 sign-ins in under four minutes. In multiple cases, the time from first targeting to privileged access in a victim’s cloud environment was under three hours. A separate password-spray campaign from the same group earlier in October peaked at about 2,900 failed sign-ins per minute.
Microsoft is careful to say that an app signing in to a tenant doesn’t prove data was taken. But the follow-on activity it documented is exactly what comes before theft: Graph reconnaissance, new app credentials, privilege changes, and in one case secrets read at scale out of Azure Key Vaults.
No human incident response team can triage a ticket, page an on-call engineer and work through a SIEM alert in three hours.
The hidden risk in enterprise app consents
We spend real money securing passwords and phone-based MFA. Yet most companies have dozens of SaaS applications holding permanent, wide-open read/write permissions through OAuth consents and enterprise app registrations. When a trusted backup vendor gets compromised, those permissions become a superhighway into your corporate data. MFA doesn’t stop it, because no human is logging in. The application is.
Where we go from here
- If you use ConnectWise Cloud Backup, follow Microsoft’s order of operations: keep the affected ConnectWise app connections disabled until ConnectWise tells you how to restore them, preserve logs and configuration before you delete anything, and search by application ID, not display name.
- Review service principal sign-ins back to at least October 15, and to October 1 if your logs go that far. Look for new credentials, role assignments, consent grants and mailbox changes.
- Audit your Entra ID enterprise apps today. Strip out inactive or abandoned app registrations. Look closely at anything holding mail, file or directory permissions.
- Treat backup and RMM tools as Tier-0 assets. If a third-party tool has keys to your tenant, give its permissions the same scrutiny you give domain administrators.
- Automate containment. If you rely on humans to spot high-velocity, cross-tenant abuse, you’re already too late. Conditional access policies and anomaly rules need to cut off tokens the moment machine-speed behavior shows up.
Where cyber insurance fits
Two parts of a cyber policy get tested by an incident like this. The first is whether a breach that starts at your vendor counts as a breach of your systems. Most modern forms cover a security failure involving systems a third party runs for you, but the definition of “computer system” is worth reading. The second is contingent business interruption: income lost when a vendor you depend on goes down, even if your own network was never touched.
Then there’s attribution. Microsoft ties this campaign to a China-state aligned group. Many cyber policies now carry war and state-backed cyber operation exclusions, and how they’re worded decides whether a nation-state label changes what gets paid. That’s a conversation to have before an incident, not during one.
The perimeter isn’t a firewall anymore. It’s your identity graph, and the attackers have automated their side of the board.